Consider whether numba could be useful

[ARF1]

It might be worth considering whether numba could help with the parallel processing effort. For factorization it probably won't help due to the need to syncronize, but maybe for other applications?

Numba seems to support just-in-time compilation and static compilation of python functions with numpy arguments into C-code and as an added bonus offers the possibility to run on CPU & GPU and releasing the GIL in the compiled code to allow effective use of python threading. This might be easier to use for many applications than using cython.parallel.

Numba Homepage
Some examples

[CarstVaartjes]

Hi Arf1!

Im a bit in pre-holiday stress atm, but will be able to look and help better in 1-2 weeks I think (from a beach with a drink ;). I've used numba (and pypy) in the past, but the last time was 2 years ago. I ran into some major issues back then (my code was too complex, breaking numba/python).
Will be interesting to try it out again; we also need to see how the numba/numexpr/cython mix will work 👍